Ground-glass opacity nodules detection and segmentation using the snake model

A ground-glass opacity (GGO) is a pulmonary shadow comprised of hazy increased attenuation with preservation of the bronchial and vascular margins in a high-resolution computed tomography (CT) image. The appearances of GGO nodules on CT images are very different from solid nodules. The current nodules segmentation method does not work well for segmenting tiny nodules and GGO nonsolid or part-solid nodules. The objective of this study is to detect and segment the GGO nonsolid or part-solid nodules. This chapter presents a semisupervised framework utilizing a nature-inspired algorithm—namely, the snake (or the active contour) model for delineating the GGO nodules. A random moving window is used for nodule detection. A region of interest will be selected based on comparison of the image density extracted from the moving window with the image density vector in a training set. We identified and tested our method on 25 GGO nodules in 55 chest CT scans. We implemented four variants of the snake model on the delineation of GGO lesion. Based on our preliminary experiment, the B-spline active contour model is more accurate and less sensitive to noise while the B-spline active Gaussian contour model is more stable and robust.